To our knowledge, this study is the first to include maternal characteristics in the development of a prediction model for childhood obesity. Applying machine learning algorithms made it possible to analyze the national panel data, with vast information on children’s healthy behaviors, maternal psychological status, and parenting style, thereby identifying risk factors for obesity in 10-year-old children. Our national data revealed the prevalence of overweight and obesity was 25% among 10-year-olds, which is similar to the prevalence found among school-aged children and adolescents in Korea.3 Since these obese children will likely become obese adults, the factors contributing to obesity development need to be investigated and managed.
The performance of our final prediction model for childhood obesity was excellent compared to those of previous prediction studies, presenting an AUC ranging from 0.73 to 0.82.17-19 Our prediction model showed higher performance than the prediction models using children’s food intake and physical activity;28 vital signs, diagnosis, and laboratory findings;17 and birth cohort data.18 These studies also applied the logistic binary classification, as we did. The findings of our study support the idea that mothers’ psychological factors need to be considered to understand childhood obesity.
It is necessary to compare our machine learning study with similar works with higher prediction findings. One study in Bangladesh developed a prediction model with 97% accuracy using data from 1100 adults.19 Another prediction model using the ensemble forest method predicted 89% of obese adults in India.16 Both prediction models used cross-sectional data and predicted risk with various features, including weight and height, which are highly correlated when it comes to obesity. Although different types of machine learning methods affect the results predicting obesity,29 our prediction model using logistic regression has an advantage in terms of determining the likelihood between risk factors and obesity development.
Our prediction models included 10 children-related (gender, eating habits, activity, and BMI at five years old) and maternal factors (education level, self-esteem, and BMI). Several machine learning algorithms used to predict obesity reported that gender, previous childhood BMI, and maternal weight and BMI were contributing factors to obesity.15–19 The gender results in our study mirror those of a study conducted in the United States, which found that more boys than girls were obese.3 These findings indicate that children’s previous weight, gender, and the BMI status of their mothers should be taken into account in fighting childhood obesity.30
Among the 10 factors included in the prediction model, demographics of the children and mothers, children’s activity levels, and maternal self-esteem were significant influencing factors. As expected, intensity and time of physical activity were also significant factors.31 We selected data concerning different intense activities─outdoor, indoor, and sedentary. In particular, time spent in indoor play that was of moderate intensity, such as playing with a toy and walking around, was a factor contributing to predicting childhood obesity in our model. This was interpreted as indicating that time spent in indoor play has a more varied range for each child compared to time spent in outdoor activities and sedentary time for five-year-old children. In the prediction model based on the intensity of physical activity, moderate-intensity—not high-intensity—activity was the most significant contributing predictor.32
This study is unique in that it includes maternal data in the development of a prediction model. Interestingly, maternal self-esteem, which has not been examined previously, was found to be a contributing predictor of developing childhood obesity. A review of the relationship between self-esteem and parenting reported that higher self-esteem predicted a higher authoritative parenting style, which is characterized by setting clear rules and expectations and solving problems together with a child. This parenting style is high in both warmth and control.22 Lopez and colleagues found that the authoritative parenting style encourages the intake of healthy food.33 However, our result demonstrated that higher self-esteem among mothers increased the likelihood of childhood obesity development. As with the results of a study on Korean parenting style, mothers’ self-esteem had a positive interrelated relationship with the warm parenting style that allowed children to do what they wanted to do not .34 This, in turn, may have influenced dietary behaviors in relation to obesity. In general, maternal self-esteem is known to influence children in many ways. Mothers’ high self-esteem mediates good coping strategies in the face of high stress, and this affects not only eating practices but also children’s healthy behaviors.35,36
Several limitations of this study should be noted. Our panel data constituted a relatively small sample size, although these data were longitudinally obtained. Careful interpretation of the findings is necessary because the data were obtained from children and families residing in South Korea, although they were meant to be representative of all Korean children’s and families’ characteristics. For future studies, we recommend utilizing a larger sample size and investigating the role of self-esteem in childhood obesity development.